ci(e2e): use ephemeral host port for e2e Postgres to stop port-collision flake

[yansun1996]

What this fixes

The native-host E2E suite intermittently fails at accounting_cluster fixture setup with:

docker: Error response from daemon: failed to bind host port 0.0.0.0:55432/tcp: address already in use

This is a top recurring flake in the E2E / native-host job (e.g. run 31138116947 errored one accounting test and went green on re-run; run 31132066804's sibling run had a leaked container hold the port for ~3h, cascading the error across every accounting test).

Root cause

Each accounting test stands up its own Postgres container on node 0, but the fixture published them all to a fixed host port (-p 55432:5432). The ~30 accounting tests run serially, each start/stop cycling Postgres. docker rm -f returns before docker-proxy actually releases the host port, so the next test's docker run intermittently collides. A leaked container can hold the port for a whole run, turning a single flake into a full-suite cascade.

Approach

Stop pinning the host port. Publish with -p 5432 so Docker allocates a free ephemeral port atomically at bind time (race-free — no window between "check free" and "bind"), then discover the assigned port via docker port and thread it into database_url. A leaked/slow-release container now holds a random port, never the one the next test needs.

SPUR_TEST_PG_PORT still pins the port when explicitly set, preserving the deterministic-port escape hatch.

Testing

Ran on a real 2-node lab: 11 accounting tests, each cycling its own Postgres container serially (the exact pattern that produced the collision on the old code). All passed; containers came up on ephemeral host ports (32780, 32795, ...) and the harness resolved each correctly. A/B check confirmed the previously-observed pool timed out failure was an unrelated stale-process artifact on the shared lab node, identical with both fixed and ephemeral ports.

[Copilot]

Pull request overview

Fixes a recurring native-host E2E flake where accounting tests intermittently fail to start Postgres due to collisions on a fixed host port, by switching to Docker-assigned ephemeral host ports and resolving the assigned port at runtime.

Changes:

• Stop pinning Postgres to a fixed host port by default; allow Docker to allocate an ephemeral port (-p 5432).
• Add logic to discover the ephemeral host port via docker port and construct database_url dynamically.
• Preserve SPUR_TEST_PG_PORT as an escape hatch to explicitly pin the host port when needed.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #590      +/-   ##
==========================================
- Coverage   76.32%   76.28%   -0.04%     
==========================================
  Files         166      166              
  Lines       65511    65511              
==========================================
- Hits        49998    49973      -25     
- Misses      15513    15538      +25     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.